Accelerometric estimates of physical activity vary unstably with data handling

Comparison of different software for processing physical activity measurements with accelerometry

Several raw-data processing software for accelerometer-measured physical activity (PA) exist, but whether results agree has not been assessed. We examined the agreement between three different software for raw accelerometer data, and associated their results with cardiovascular risk. A cross-sectional analysis conducted between 2014 and 2017 in 2693 adults (53.4% female, 45-86 years) living in Lausanne, Switzerland was used. Participants wore the wrist-worn GENEActive accelerometer for 14 days. Data was processed with the GENEActiv manufacturer software, the Pampro package in Python and the GGIR package in R. For the latter, two sets of thresholds "White" and "MRC" defining levels of PA and two versions (1.5-9 and 1.11-1) for the "MRC" threshold were used. Cardiovascular risk was assessed using the SCORE risk score. Time spent (mins/day) in stationary, light, moderate and vigorous PA ranged from 633 (GGIR-MRC) to 1147 (Pampro); 93 (GGIR-White) to 196 (GGIR-MRC); 19 (GGIR-White) to 161 (GENEActiv) and 1 (GENEActiv) to 26 (Pampro), respectively. Spearman correlations between results ranged between 0.317 and 0.995, while concordance coefficients ranged between 0.035 and 0.968. With some exceptions, the line of perfect agreement was not in the 95% confidence interval of the Bland-Altman plots. Compliance to PA guidelines varied considerably: 99.8%, 98.7%, 76.3%, 72.6% and 50.2% for Pampro, GENEActiv, GGIR-MRC v.1.11-1, GGIR-MRC v.1.4-9 and GGIR-White, respectively. Cardiovascular risk decreased with increasing time spent in PA across most software packages. We found large differences in PA estimation between software and thresholds used, which makes comparability between studies challenging.

A Standardized Buffalo Concussion Treadmill Test Following Sport-Related Concussion in Youth: Do ActiGraph Algorithms Matter?

CONTEXT

Current recovery guidelines following sport-related concussion (SRC) include 24-48 hours of rest followed by gradual return-to-activity with heart rate (HR) maintained below symptom threshold. Additionally, the monitoring of physical activity (PA) post-SRC using ActiGraph accelerometers can provide further objective insight on amounts of activity associated with recovery trajectories. Cut-point algorithms for these devices allow minute-by-minute PA to be classified into intensity domains; however, studies have shown different algorithms employed on the same healthy participant dataset can produce varying classifications.

OBJECTIVE

To identify the most physiologically appropriate cut-point algorithm (Evenson or Romanzini) to analyze ActiGraph data in concussed youth with comparisons to HR response on the Buffalo Concussion Treadmill Test (BCTT).

DESIGN

Prospective cohort study.

SETTING

Sport-concussion clinic within a university setting.

PATIENTS OR OTHER PARTICIPANTS

Eleven high-school students (5 male, 6 female; median [range] age =16 years [15-17], height = 177.8 cm [157.5-198.1], weight = 67 kg [52-98], body mass index = 22 kg/m2 [17-31]) involved in high-risk sport who sustained a physician diagnosed SRC.

MAIN OUTCOME MEASURE(S)

Evenson and Romanzini algorithm PA intensity domains via ActiGraph data and HR during the BCTT.

RESULTS

There were differences in moderate (P = .001) and vigorous (P = .002) intensities between algorithms, but no difference in light (P = .548). Evenson classified most of the time as moderate intensity (57.03% [0.00-94.12%]), whereas Romanzini classified virtually all PA as vigorous (88.25% [2.94-97.06%]). PA based on HR (stages 1-7: 20-39% HR reserve (HRR), stages 8-13: 40-59% HRR, stages 14 and above: 60-85% HRR) indicated the BCTT primarily involves light-to-moderate intensity, and therefore is better represented by the Evenson algorithm.

CONCLUSIONS

The Evenson algorithm better characterizes the HR response during a standardized exercise test in concussed individuals and therefore should be used to analyze ActiGraph PA data in a concussed paediatric population.

Show cards of the Global Physical Activity Questionnaire (GPAQ) - do they impact validity? A crossover study

Background

The Global Physical Activity Questionnaire (GPAQ) is applied internationally as a tool to assess the level of physical activity. The GPAQ was designed as an interview, including the use of show cards, which visualise activities of moderate and intensive physical activity and support the distinction between these intensities. The self-administered version of the GPAQ is used in the application-oriented research for reasons of economy and practicality. However, the use of show cards often remains unknown.

The aim of the present study was to examine differences in validity between two self-administered versions of the GPAQ with and without show cards.

Methods

In this crossover study, two groups (n = 54; 57.4% female; 28.3 ± 12.2 years) received the GPAQ with or without show cards after 7 days and the respective other version after additional 7 days. For validation, all participants wore an accelerometer (ActiGraph GT3X+) on all 14 days.

Differences between GPAQ versions and accelerometer data were compared by Wilcoxon signed rank test. Additionally, Spearman analyses and Bland-Altman plots were calculated.

Results

No statistically significant difference between the GPAQ versions could be found in regard to the accuracy of physical activity assessment (p > 0.05).

Both GPAQ versions show similar correlation coefficients for vigorous physical activity (rho = 0.31–0.42) and sedentary behaviour (rho = 0.29–0.32). No statistically significant correlation was found for physical activity of moderate intensity. The Bland-Altman plots support these results, as both GPAQ versions have the same trends in terms of overestimation and underestimation of physical activity.

Conclusion

The use of show cards had no significant impact on questionnaire validity. Therefore, both GPAQ versions can be applied interchangeably. Nevertheless the exact description of application of the GPAQ is desirable in terms of reproducibility and transparent scientific research.

Physical activity levels in adults and elderly from triaxial and uniaxial accelerometry. The Tromsø Study

Introduction

Surveillance of physical activity at the population level increases the knowledge on levels and trends of physical activity, which may support public health initiatives to promote physical activity. Physical activity assessed by accelerometry is challenged by varying data processing procedures, which influences the outcome. We aimed to describe the levels and prevalence estimates of physical activity, and to examine how triaxial and uniaxial accelerometry data influences these estimates, in a large population-based cohort of Norwegian adults.

Methods

This cross-sectional study included 5918 women and men aged 40–84 years who participated in the seventh wave of the Tromsø Study (2015–16). The participants wore an ActiGraph wGT3X-BT accelerometer attached to the hip for 24 hours per day over seven consecutive days. Accelerometry variables were expressed as volume (counts·minute^-1 and steps·day^-1) and as minutes per day in sedentary, light physical activity and moderate and vigorous physical activity (MVPA).

Results

From triaxial accelerometry data, 22% (95% confidence interval (CI): 21–23%) of the participants fulfilled the current global recommendations for physical activity (≥150 minutes of MVPA per week in ≥10-minute bouts), while 70% (95% CI: 69–71%) accumulated ≥150 minutes of non-bouted MVPA per week. When analysing uniaxial data, 18% fulfilled the current recommendations (i.e. 20% difference compared with triaxial data), and 55% (95% CI: 53–56%) accumulated ≥150 minutes of non-bouted MVPA per week. We observed approximately 100 less minutes of sedentary time and 90 minutes more of light physical activity from triaxial data compared with uniaxial data (p<0.001).

Conclusion

The prevalence estimates of sufficiently active adults and elderly are more than three times higher (22% vs. 70%) when comparing triaxial bouted and non-bouted MVPA. Physical activity estimates are highly dependent on accelerometry data processing criteria and on different definitions of physical activity recommendations, which may influence prevalence estimates and tracking of physical activity patterns over time.

Computer Vision-Based Unobtrusive Physical Activity Monitoring in School by Room-Level Physical Activity Estimation: A Method Proposition

As sedentary lifestyles and childhood obesity are becoming more prevalent, research in the field of physical activity (PA) has gained much momentum. Monitoring the PA of children and adolescents is crucial for ascertaining and understanding the phenomena that facilitate and hinder PA in order to develop effective interventions for promoting physically active habits. Popular individual-level measures are sensitive to social desirability bias and subject reactivity. Intrusiveness of these methods, especially when studying children, also limits the possible duration of monitoring and assumes strict submission to human research ethics requirements and vigilance in personal data protection. Meanwhile, growth in computational capacity has enabled computer vision researchers to successfully use deep learning algorithms for real-time behaviour analysis such as action recognition. This work analyzes the weaknesses of existing methods used in PA research; gives an overview of relevant advances in video-based action recognition methods; and proposes the outline of a novel action intensity classifier utilizing sensor-supervised learning for estimating ambient PA. The proposed method, if applied as a distributed privacy-preserving sensor system, is argued to be useful for monitoring the spatio-temporal distribution of PA in schools over long periods and assessing the efficiency of school-based PA interventions.

Uni- and triaxial accelerometric signals agree during daily routine, but show differences between sports

Accelerometers objectively monitor physical activity, and ongoing research suggests they can also detect patterns of body movement. However, different types of signal (uniaxial, captured by older studies, vs. the newer triaxial) and or/device (validated Actigraph used by older studies, vs. others) may lead to incomparability of results from different time periods. Standardization is desirable. We establish whether uniaxial signals adequately monitor routine activity, and whether triaxial accelerometry can detect sport-specific variations in movement pattern. 1402 adolescents wore triaxial Actigraphs (GT3X) for one week and diaried sport. Uni- and triaxial counts per minute were compared across the week and between over 30 different sports. Across the whole recording period 95% of variance in triaxial counts was explained by the vertical axis (5th percentile for R², 91%). Sport made up a small fraction of daily routine, but differences were visible: even when total acceleration was comparable, little was vertical in horizontal movements, such as ice skating (uniaxial counts 41% of triaxial) compared to complex movements (taekwondo, 55%) or ambulation (soccer, 69%). Triaxial accelerometry captured differences in movement pattern between sports, but so little time was spent in sport that, across the whole day, uni- and triaxial signals correlated closely. This indicates that, with certain limitations, uniaxial accelerometric measures of routine activity from older studies can be feasibly compared to triaxial measures from newer studies. Comparison of new studies based on raw accelerations to older studies based on proprietary devices and measures (epochs, counts) will require additional efforts which are not addressed in this paper.

Physical activity and disability measures in chronic non-specific low back pain: a study of responsiveness

Objectives:

To compare the responsiveness of disability measures with physical activity measures in patients with chronic low back pain (CLBP) undergoing a course of physical therapy treatment.

Design:

This is a prospective cohort study with two-month follow-up.

Subjects:

A total of 106 patients presenting with non-specific CLBP of more than three months duration were recruited.

Main measures:

Disability measures investigated were Quebec Back Pain Disability Scale and Roland Morris Disability Questionnaire. Physical activity measures analyzed include the Baecke Habitual Physical Activity Questionnaire and objective measures derived from an accelerometer (i.e. total time spent in moderate-to-vigorous and light physical activity, number of steps and counts per minute). Disability and physical activity measures were collected at the baseline and after eight weeks of treatment. For the responsiveness analyses, effect size (ES) and standardized response mean (SRM) were calculated. Correlations between the change in disability and physical activity measures were calculated.

Results:

Responsiveness for disability measures was considered to be large with ESs ranging from −1.03 to −1.45 and SRMs ranging from −0.99 to −1.34, whereas all physical activity measures showed values lower than 0.20. Changes in disability measures did not correlate with changes in physical activity measures (correlation coefficients ranged from −0.10 to 0.09).

Conclusion:

Disability measures were responsive after a course of physical therapy treatment in patients with CLBP. The lack of responsiveness in the physical activity measures might be due to the inability of these measures to detect change over time or the use of an intervention not designed to increase physical activity levels.